- Neuron pair programming panel (right column, 300px): full conversation interface
with code-block rendering, "Insert at cursor" for AI suggestions, quick actions
(explain selection, find activations, analyze types), thinking indicator
- Live activation preview: floating inline widget when cursor is inside
`activate TypeName where "..."` expressions; queries /api/lsp/activate-preview
- Type graph activation highlighting: `activate` statements cause named types to
pulse with purple glow rings; animation loop runs while activations are present
- Knowledge Graph Explorer: new bottom panel tab; search the Engram DB for semantic
concepts, insert activate expressions, ask Neuron about nodes
- Semantic autocomplete: completions with score >= 0.8 get "⟁ semantic" prefix and
priority boost in the completion list
- Backend: /api/reason extended to support three modes: "hypothesis" (existing),
"pair" (proxies to soma /v1/chat/completions with full file context + system
prompt), "knowledge-search" (proxies to Engram DB search endpoint)
- Backend: new GET /api/lsp/activate-preview endpoint queries Engram for nodes
matching a partial activate expression
- Type graph node click now navigates editor to the type definition
- Resolved merge conflicts in engram-lang/crates (ast.rs, parser.rs, checker.rs,
types.rs): took union of HEAD and worktree-agent branches
- Register print/println/log/print_err in TypeEnv::with_builtins() with polymorphic (Unknown) param type so any value type is accepted without spurious warnings
- Add StructLit expr to AST + parser (uppercase-IDENT { ... } syntax), BuildStruct bytecode instruction + Struct value to runtime
- Add type_params to FnDef AST node and TypeParam variant to TypeExpr; parser parses <T, E> generics; type checker treats TypeParam as Unknown (universal type)
- Rewrite interpreter to support user-defined function calls via call stack (Frame + return_ip); dispatch_builtin handles print/println/log/print_err/__build_list__
- Fix engram_activate_search to unwrap { results: [...] } response envelope from /search; use std::net for sync HTTP to avoid reqwest dependency
- Add run-file command to CLI for single-file execution without el.toml
- Fix worktree engram-crypto path dep
Axum HTTP server (port 7771) serving a single-page IDE with CodeMirror 6
syntax highlighting for engram-lang, a force-directed type graph visualizer,
LSP (completions, hover, diagnostics), SSE-streamed build/run output, a
plugin host with five first-party plugins, and a reasoning panel that proxies
to engram-server. 28 tests across three crates, zero warnings.
- New crate el-test: test discovery, in-memory graph seeding, assertion evaluator, TestRunner, TestReport with human/JSON/JUnit XML output
- New keywords: test, seed, assert, target — fully integrated into lexer, parser, codegen, type-checker
- Parser extensions: TestDef, SeedStmt, Assert AST nodes; seed blocks handle type: as field name (keyword-as-ident in seed context)
- Debugger: DebugEvent, Debugger, StepMode, StackFrame in el-compiler — breakpoints, step-over, step-into, step-out
- CLI: el test-file <file.el> runs tests; el test integrates with project; el debug attaches debugger; --output json|junit for CI
- 52 new tests in el-test covering discovery, graph seeding, assertion evaluation, pass/fail/error/skip, report generation, JUnit XML
- Example: examples/hello-project/src/tests.el — 6 unit tests pass, 1 e2e test correctly skipped without ENGRAM_URL
Memory is not stored and retrieved — it is activated and propagated.
Implements the spreading activation model with salience decay, typed edges,
four memory tiers, and flat cosine vector search over a sled embedded store.
Will Anderson